粒子群算法优化神经网络结构的研究  

Particle Swarm Algorithm Used for Optimizing Neural Network Structure

作　　者：田雨波[1] 潘朋朋[1] 

机构地区：[1]江苏科技大学电子信息学院,江苏镇江212003

出　　处：《现代电子技术》2011年第4期110-112,共3页Modern Electronics Technique

基　　金：江苏省高校自然科学基础研究项目(07K5B510032)

摘　　要：针对BP神经网络初始权阈值确定的随机性和隐含层节点数的不确定性,通过利用十进制粒子群优化算法(DePSO)和二进制粒子群优化算法(BiPSO),同时优化神经网络的初始权阈值和结构。通过粒子群优化算法首先确定一个较好的搜索空间,然后在这个解空间里利用BP算法对网络进行训练和学习,搜索出最优解。通过函数拟合数值实验对该模型来进行训练和测试,相比其他算法,该模型可以获得较高的预测精度,结果表明该方法是可行的。Aiming at the random of initial weight threshold and uncertainty of hidden layer nodes for BP neural network, the decimal and binary particle swarm optimization algorithm were used to optimize the initial weight threshold and structure of the neural network. A better search space was confirmed by particle swarm optimization firstly, and then the network is trained and learned in the solution space by BP algorithm, thereby, the optimal solution can be searched. The function fitting numerical experiment was used to train and test this model. A higher forecast accuracy can be gained. The result shows that the way is feasible.

关 键 词：粒子群 神经网络 隐含层节点数 函数拟合 

分 类 号：TN911-34[电子电信—通信与信息系统] TP183[电子电信—信息与通信工程]